In a significant development for the AI community, German startup Black Forest Labs has announced its launch and the release of FLUX.1, a suite of cutting-edge text-to-image AI models. Founded by researchers who played key roles in developing Stable Diffusion and inventing the latent diffusion technique, Black Forest Labs aims to push the boundaries of generative AI for images and videos.

FLUX.1 comes in three variants:

1. FLUX.1 [pro]: The flagship model, offering state-of-the-art performance in image generation with superior prompt following, visual quality, image detail, and output diversity. 2. FLUX.1 [dev]: An open-weight, guidance-distilled model for non-commercial applications, available on HuggingFace. 3. FLUX.1 [schnell]: A faster open-weights version designed for local development and personal use, also available on HuggingFace.

The company claims that FLUX.1 outperforms existing options like Midjourney and DALL-E in areas such as image quality and adherence to text prompts. Initial tests suggest that the outputs of the two higher-end FLUX.1 models are comparable to OpenAI's DALL-E 3 in prompt fidelity, with photorealism close to Midjourney 6.

FLUX.1 models use a "hybrid architecture" combining transformer and diffusion techniques, scaled up to 12 billion parameters. The company improves on previous diffusion models by incorporating flow matching and other optimizations. Notably, FLUX.1 demonstrates competence in generating human hands, a historical weak spot for earlier image-synthesis models.

Black Forest Labs has secured $31 million in Series Seed funding, led by Andreessen Horowitz, with additional investments from General Catalyst and MätchVC. The company has also brought on high-profile advisers, including former Disney President Michael Ovitz and AI researcher Matthias Bethge.

While currently focused on text-to-image generation, Black Forest Labs plans to expand into video generation next. They claim that FLUX.1 will serve as the foundation for a new text-to-video model in development, aiming to compete with offerings from OpenAI, Runway, and Kuaishou.

The company emphasizes its commitment to making AI models accessible to a wide audience, aiming to foster innovation and increase transparency. However, questions remain about the source of the training data used to develop FLUX.1, with speculation that it may involve large-scale web scraping similar to methods used by Stability AI.

For those interested in trying FLUX.1, the models are available through various platforms:

- FLUX.1 [pro] is accessible via the company's API, Replicate, and fal.ai. - FLUX.1 [dev] and [schnell] weights are available on HuggingFace, with inference code on GitHub and integration for ComfyUI.

As Black Forest Labs continues to develop its technology, the impact of FLUX.1 on the AI image generation landscape remains to be seen. However, with its strong technical foundation and ambitious plans, the company is positioning itself as a significant player in the rapidly evolving field of generative AI.